透過您的圖書館登入
IP:18.119.162.247
  • 期刊

結合類神經網路與遺傳演算法於翡翠水庫最佳控制優養化之研究

A Study of Combining the Neural Networks with Genetic Algorithms to the Optimal Control of Eutrophication in Feitsui Reservoir

摘要


近來水庫優養化已成爲嚴重的水質污染問題,影響水庫優養有許多因素,包括水庫的大小、深度和形狀,以及日照的強度、氣溫和營發鹽等。本研究考量污染源之時變性,企圖以類神經網路建立水質預測模式。此模式是根據六個重要關係因子來預測翡翠水庫之水質,其中包括北勢溪總磷濃度(μg/L)、兩條支流鰱魚窟溪總磷濃度(μg/L)、金瓜寮溪總磷濃度(μg/L)、日最高降雨量(mm)、月平均降雨量(mm)、與水庫的淨入流量(CMSD)(入與出流量之差值)等。由於這些因子存在非線性關係,因此本研究利用類神經網路結合遺傳演算法來探討降低磷自支流進入水庫中的可行性,以維持水庫良好之水質狀態。

並列摘要


Eutrophication has been a serious problem which results in the pollution of water quality in reservoirs recently. A lot of factors including the size, depth and shape of a reservoir as well as the intensity of sun light, could affect nutrients concentration in reservoirs. In this study, we consider that the source pollution is unstable and attempt to use artificial neural networks (ANNs) to construct the water quality forecast models. This model is based on data from nutrient loads (which include the inflows from a main creek and two tributaries), average and maximum rainfall in the watershed, and net inflow (difference between inflow and outflow) to forecast the water quality in Feitsui Reservoir. These models use variant factors to predict the dynamic nutrient concentration. Owing to these factors exist non-linear relationships; ANNs were hybridized with genetic algorithms (GAs) to explore the feasibility of controlling phosphorus loads into the reservoir and maintaining a satisfied situation of water quality.

延伸閱讀